Reproduce “HyCo: Hybrid Drafting for Cost-Efficient Mobile-Cloud Speculative Decoding” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hanbin Bae, Heejin Choi, Joohyung Lee, Byoung Jin Choi, Sangjun Park, Hoon-Young Cho
HyCo reallocates drafting work in device–cloud speculative decoding: the verifier’s multi-token-prediction head builds a token chain on the phone, while one SLM pass adds alternatives. In Spec-Bench tests, HyCo saved 9 percentage points more server time than SpecEdge-ε under the same acceptance rule and delivered 35.0 tokens per second versus 18.9. This does not establish equivalent quality or loaded-service capacity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.